Optimizely Opal instruction agents overview

  • Updated

Although Optimizely Opal has many features that do not require configuration, you must instruct Optimizely Opal on approaching certain topics to tailor its output. This gives it some personality and increases its awareness of your brand tone of voice and your competitors.

Instruction agents

Instruction agents are the core of Opal's dynamic instruction-following capabilities. They are sets of predefined rules and steps that tell Opal how to complete a task when asked or triggered by a user or process, including what to do, how to do it, and when to do it. These repeatable structures let you create reusable workflows and automate common tasks. A simple form of an instruction agent could have your company's brand guidelines, ensuring any content generated by Opal is on-brand.

You can invoke instruction agents directly or indirectly through user intent, providing flexibility and control over Opal's behaviour. For example, a user might directly request a campaign brief and trigger the corresponding instruction agent, or they may ask a question like How can I plan my next product launch?, leading Opal to activate the same agent.

Components of an instruction agent

Each instruction agent contains the following components:

  1. Name – A short, descriptive identifier for the agent, aiding in easy recognition.
  2. Details – Comprehensive instructions outlining the agent's behavior, including objectives, execution steps, response formats, and any constraints.
  3. Where to Use – Specifies the product and instance contexts where the agent is applicable.
  4. When to Use – Defines the triggers or conditions under which the agent should be activated.
  5. Active Status – Indicates whether the agent is currently enabled and operational.
After configuring an instruction agent, ensure you activate it. An inactive agent will not respond to triggers or perform tasks, regardless of its configuration.

Default instructions

Opal includes default instructions for generating content across formats, such as blog posts, social copy, suggestions, and help responses. These instructions act as reusable prompt frameworks, which you can tailor to your brand voice or content goals.

The prebuilt instructions are turned off by default. See Customize Opal for instructions on enabling them. 

Prebuilt instruction agents

Opal includes the following instruction agents that you can modify:

Optimizely Content Marketing Platform

  • Campaign Brief (None) – Create a campaign brief in a defined structure.
  • Industry Marketer (@industrialise, @industrialize, or @industry) – Provide a URL and create an industry-targeted variation.
  • Keyword Research (@keywords) – Give a targeted keyword and see suggestions and recommendations.
  • Marketing Planning Meeting (@meetinglaunch) – Create a campaign and brief based on the content of a video upload.
  • Marketing Researcher (@research) – Research and define a campaign marketing strategy that forms the campaign brief.
  • Task Brief (None) – Create a task brief in a defined structure.
  • Tone of Voice Sample (None) – Provide instructions on your tone of voice.
  • Video Transcription (@transcribe) – Provide transcriptions of uploaded videos. Supports text, subtitle, and translation as outputs.

Optimizely Experimentation

  • Experiment Plan (@ExperimentPlan) – Get feedback on a test plan for Optimizely Experimentation. This agent is not customizable.
  • Personalization Advisor (@PersonalizationAdvisor) – Use this agent to design and measure high-value personalization opportunities.

Custom instructions

Admins can create custom instruction agents, see Create and edit instructions.